Abstract
Sufficient manpower on ship bridges plays an important role in ship navigation and watchkeeping safety. Effectively counting the number of watchkeeping officers and evaluating compliance with manning rules can prevent many ship accidents caused by fatigue and poor lookouts. Current solutions mostly rely on surveillance cameras and watchmen to determine whether to report to the shipmaster, a system that requires extra human effort and raises concerns regarding officers’ privacy. In this paper, we utilize WiFi infrastructure in a ship bridge and propose a WiFi-signal-based watchkeeping officer counting method based on fine-grained channel state information (CSI) instead of using cameras. To achieve this goal, we first adopt the Pearson correlation coefficient to evaluate the sensitivity of CSI and extract the CSI subcarriers that are most sensitive to human activities, rather than environmental changes. Then, an expansion matrix algorithm is used to estimate the number of people in the line-of-sight WiFi link using the extracted CSI subcarriers, and a multi-link fusing scheme is proposed to determine the total number of officers in the bridge by using multiple WiFi devices. Finally, after obtaining the number of current watchkeeping officers, combined with the current spatio-temporal properties of the ship, the system determines the number of safe navigation personnel. If the number of people detected does not meet the requirements, the system will issue an alarm. We have conducted extensive experiments in a real-world passenger ship. The experimental results show that after introducing the sensitive subcarrier extraction module, the accuracy of counting the number of officers increased by at least 17%. When the total number was less than 4 or equal to 1, the system achieved an average accuracy of 88% and 94%, respectively.
Funding Information
  • National Natural Science Foundation of China (51979216)
  • Natural Science Foundation of Hubei province, China (2021CFA001)
  • Green Intelligent Inland Ship Innovation Programme

This publication has 32 references indexed in Scilit: